Skip to contents

geo_epidemic_simulate() runs a stochastic, spatially explicit SIR (susceptible-infected-recovered) model on a grid of host populations, for example a wildlife disease such as chytridiomycosis in frogs, a plant pathogen in a forest, or an outbreak among villages. Each cell holds hosts individuals. In each time step a susceptible host in cell \(j\) is infected with probability \(1 - e^{-\lambda_j}\), $$\lambda_j = \beta \left[(1 - m) \frac{I_j}{N_j} + m \sum_i K(d_{ij}) \frac{I_i}{N_i}\right],$$ where \(\beta\) is the transmission rate, \(m\) the share of contacts made outside the home cell (coupling) and \(K\) the normalised dispersal kernel of geo_shift_simulate() (exponential or fat-tailed 2Dt, with optional resistance). Infected hosts recover with probability \(1 - e^{-\gamma}\) per step, and recovered hosts lose immunity with probability \(1 - e^{-\omega}\) (omega = 0: lasting immunity). Time runs in discrete steps, so an infected host stays infectious for \(1 / (1 - e^{-\gamma})\) steps on average (close to \(1/\gamma\) when \(\gamma\) is small), and in a single well-mixed population the basic reproduction number is \(R_0 = \beta / (1 - e^{-\gamma})\). The expected final share infected \(z\) then solves \(z = 1 - e^{-R_0 z}\); the tests check the simulation against this.

Usage

geo_epidemic_simulate(
  hosts,
  initial,
  beta,
  gamma,
  omega = 0,
  coupling = 0.1,
  dispersal_mean,
  times,
  kernel = c("exponential", "2Dt"),
  kernel_shape = 1,
  max_distance = NULL,
  resistance = NULL,
  initial_count = 1,
  nsim = 100,
  seed = NULL,
  threads = 0
)

Arguments

hosts

SpatRaster of the number of hosts per cell (rounded to whole numbers; NA or 0 = no hosts).

initial

Initially infected hosts: a SpatRaster of counts, or an sf/SpatVector of points, each infecting initial_count hosts in its cell.

beta

Transmission rate per time step.

gamma

Recovery rate per time step (mean infectious period \(1 / (1 - e^{-\gamma})\) steps).

omega

Rate of loss of immunity per step (0 = lasting immunity).

coupling

Share of contacts outside the home cell, 0 to 1.

dispersal_mean

Mean distance of contacts outside the cell, in metres.

times

Time steps at which to record the maps (increasing integers).

kernel

"exponential" or "2Dt" (fat-tailed).

kernel_shape

Shape \(p\) of the 2Dt kernel (above 0.5; smaller is fatter-tailed).

max_distance

Largest dispersal distance in metres (kernel truncation); default five times the mean for the exponential kernel and ten times for the 2Dt kernel.

resistance

Optional SpatRaster of movement resistance (1 = baseline; NA, 0 or Inf = barrier).

initial_count

Hosts infected at each starting point.

nsim

Number of stochastic simulations.

seed

Random seed.

threads

Number of threads (0 = OpenMP default).

Value

A list of class tmr_epidemic:

  • prevalence: SpatRaster, mean share of hosts infected at each of times;

  • reached: SpatRaster, probability that the disease has reached each cell by each of times;

  • curve: tibble of time and mean number infected per step;

  • attack: share of all hosts ever infected, per simulation;

  • settings.

References

Keeling, M. J. and Rohani, P. (2008). Modeling Infectious Diseases in Humans and Animals. Princeton University Press.

Examples

# \donttest{
if (requireNamespace("terra", quietly = TRUE)) {
  r <- terra::rast(
    nrows = 30, ncols = 30, xmin = 0, xmax = 30000,
    ymin = 0, ymax = 30000, crs = "EPSG:32755"
  )
  hosts <- terra::init(r, 200)
  seed <- terra::init(r, 0)
  seed[15, 15] <- 5
  ep <- geo_epidemic_simulate(hosts, seed,
    beta = 0.4, gamma = 0.1, coupling = 0.3,
    dispersal_mean = 2000, times = c(20, 40, 80), nsim = 10, seed = 1
  )
  terra::plot(ep$reached)
  mean(ep$attack)
}

#> [1] 0.9653506
# }